Extended Data Fig. 7: Raw and smoothed maps of pollen-based estimates of millennial temperature variability, and location of the records contributing to each spectral bin.

a. Same as Fig. 2a, but for the individual sites. b. Same as a but after applying a 300-km Gaussian smoothing. The opacity is proportional to the effective number of records contributing to the smoothed value, that is the sum of the Gaussian weights, and is fully opaque when 5 or more effective records were available. c. Shown are the location of individual pollen records that were considered in the analysis. The colours correspond to those in Fig.3, indicating which records are included in the binning of each spectrum. d. Shown are the grid points of the instrumental dataset that are near pollen records. The colours also indicate the corresponding bins as in a and Fig. 3. Each of the 8 non-overlapping bins contain 78 grid points based on the sub-decadal variability (see Methods Sub-Decadal Variability Binning).